betaFeedback

WorksPhysics and Astronomy

A Mixed-Metric Two-Field Framework for Turbulence: Emergent Stress Anisotropy and Wall Asymptotics from a Single Scalar

Abstract

In our previous work~\cite{SanchisAgudoVinuesa2025PRL}, we argued that viscous dissipation in turbulence can be understood as the macroscopic imprint of microscopic path uncertainty, and showed that a kernel variance field s(y)s(y) constrained by a balance condition yields both the Kolmogorov scales and the logarithmic law of the wall from a single stochastic principle. In the present work we promote ss to a dynamical field s(x,t)s(\bm{x},t) with units of kinematic viscosity and develop a two-field framework in which the velocity \ve and an intermittency (or stochastic diffusivity) field ss evolve in a coupled way. The effective viscosity is νeff=ν0+sν_{\mathrm{eff}}=ν_0+s, but the stress tensor is generalized to include a non-linear closure driven by the commutator of strain and rotation, [S,Ω][\bm{S}, \bmΩ], capturing emergent anisotropy. The evolution of ss is defined as a mixed-metric gradient flow: a Wasserstein-2 gradient flow for morphology, \Div(s\grad s), combined with a local L2L^2 gradient flow driven by an objective coupling term qq. The coupling is decomposed as q=qprod−qrelaxq=q_{\mathrm{prod}}-q_{\mathrm{relax}}, where production is driven by a vortex-stretching invariant, I=∥Sω∥2\mathcal{I} = \|\bm{S}\boldsymbolω\|^2. This choice ensures that production vanishes identically in strictly two-dimensional flows. We show that, under standard assumptions of constant stress, high Reynolds number and overlap-layer scale invariance, the only scale-invariant overlap-layer solution of the mixed-metric equation is s(y)∝ys(y)\propto y, which recovers the logarithmic velocity profile. Thus the same mixed-metric equation organizes both wall-resolved and wall-modeled asymptotics within a single, energetically constrained framework.

Introduction

A significant fraction of the challenge in predicting turbulent flows is related to the behavior of the near-wall region. In our previous work [1] we took a geometric viewpoint and asked: what if viscous dissipation is the macroscopic imprint of a crowd trying to cross a plaza? Imagine crossing a “grand plaza” from A to B. In the empty plaza you walk along a straight geodesic, the world of least action: this is Arnold’s ideal fluid [2]. In a crowded plaza, your intention (mean path) is still approximately geodesic, but your actual trajectory jitters as you dodge others. You travel through a cloud of paths, and the trip takes longer on average. In our previous work [1] we encoded that cloud by a kernel variance field and showed that constraining this exploratory field near a wall is enough to recover two pillars of turbulence theory: the Kolmogorov scales (velocity, time and length) and the logarithmic law of the wall.

The present paper asks a sharper question: what is ss as a physical field, and can we write an evolution equation for it that is both constrained by geometry and useful for modeling?

From algebraic closures to a dynamical ss-field. Classical models treat unresolved turbulence through an effective viscosity, whether as an algebraic eddy viscosity

νt=(CSΔ)2Korνt∼ℓm2∣U′∣\nu_t=(C_S\Delta)^2K \quad\text{or} \quad\nu_t\sim\ell_m^2\lvert U'\rvert

in LES and mixing-length theories [3, 4, 5, 6], or through more elaborate two-equation closures such as kk–ϵ\epsilon or kk–ω\omega [7, 8]. In all cases, the effective viscosity is built as a functional of the resolved velocity gradient and a few extra scalars, chosen to satisfy symmetries and match data, but not derived from an underlying evolution principle.

Our previous work [1] suggested a different starting point: viscosity as a consequence of microscopic path uncertainty. In that picture, ss is not just a tuning function but the field that quantifies how much trajectories explore around the mean flow. Here we adopt that interpretation and promote ss to a space-time field s(x,t)s(x,t) that co-evolves with the velocity:

  • the velocity v\boldsymbol{v} describes the exploit dynamics, i.e. the organized mean motion across the plaza;

  • the scalar field ss describes the explore dynamics, i.e. the local intensity of microscopic wandering that gives rise to dissipation.

Throughout this paper we assign ss the units of kinematic viscosity (L2/TL^2/T) and interpret it as a stochastic diffusivity (a variance rate in path space). The effective viscosity is then

νeff(x,t)=ν0+s(x,t),(1)\nu_{\mathrm{eff}}(x,t)=\nu_0+s(x,t), \tag*{(1)}

with ν0\nu_0 being the molecular contribution and ss the turbulent one. Relation to the kernel variance in our previous work. In our previous work [1], the symbol s(y)s(y) appeared as a variance scale with units of length squared. Here s(x,t)s(x,t) has units of L2/TL^2/T, because it enters additively in νeff=ν0+s\nu_{\mathrm{eff}}=\nu_0+s and plays the role of a stochastic diffusivity. The two uses are consistent if one introduces a small kernel time scale Δt∗\Delta t_* and writes the variance of the underlying Brownian exploration as

σ2(y)∼2νeff(y)Δt∗≡2[ν0+s(y)]Δt∗.(2)\sigma^2(y) \sim2\nu_{\mathrm{eff}}(y)\Delta t_* \equiv2[\nu_0+s(y)]\Delta t_*. \tag*{(2)}

In this coarse-grained picture, the “variance profile” of our previous work is σ2(y)\sigma^2(y), whereas the present intermittency field is

s(y)=νeff(y)−ν0.(3)s(y)=\nu_{\mathrm{eff}}(y)-\nu_0. \tag*{(3)}

Up to the constant factor 2Δt∗2\Delta t_*, both encode the same wall-normal structure of path uncertainty: in our previous work we emphasized the (dimensionful) variance, here we take the corresponding variance rate and use it as a dynamical field. For this reason we reuse the symbol ss in the two-field framework, implicitly absorbing Δt∗\Delta t_* into its definition.

The central question of this paper is how to define the evolution of ss in a way that is:

  1. variationally structured, so that some part of its dynamics is fixed by a gradient-flow principle rather than ad hoc;

  2. objectively coupled to the resolved velocity gradient, through an invariant that detects genuinely three-dimensional stretching;

  3. compatible with wall physics, in the sense that it recovers both the viscous sublayer and the logarithmic overlap layer as asymptotic limits.

Summary of the framework. We propose a two-field framework in which:

  • the momentum equation is closed by νeff=ν0+s\nu_{\mathrm{eff}}=\nu_0+s plus an emergent non-linear stress term that accounts for rotational anisotropy,

  • the field ss evolves according to a mixed-metric gradient flow: a Wasserstein-2 gradient flow of a simple “morphology” energy, yielding the degenerate diffusion ∇⋅(s∇s)\nabla\cdot(s\nabla s) familiar from porous-medium theory [9, 10, 11], plus a local L2L^2 gradient flow of a reaction potential that encodes production and relaxation.

We are explicit that this is a modeling framework structured by axioms, not a full derivation from the Schrödinger-Bridge formalism of our previous work [1]: the SB perspective motivates the existence and additivity of a scalar diffusivity field, but the concrete choices for the evolution of ss are specified below as physically guided modeling assumptions.

A key structural element is the invariant driving production. We argue that the natural choice is the squared norm of the vortex-stretching vector,

I(S,ω)=∥Sω∥2,(4)\mathcal{I}(S,\omega)=\lVert S\omega\rVert^2, \tag*{(4)}

where SS is the strain-rate tensor and ω=∇×v\omega=\nabla\times v the vorticity. This quantity measures how strongly the flow stretches and tilts vortical structures, vanishes identically in strictly two-dimensional flows, and has the correct dimension to enter the source term of the ss-equation.

The rest of the paper is organized as follows. In Sec. II we formalize the two-field setup and notation. In Sec. III we derive the mixed-metric gradient flow for ss and specify the production and relaxation terms. In Sec. IV we analyze the wall asymptotics and show that the same equation yields s∼y2s\sim y^2 as y→0y\to0 and s∝ys\propto y in the overlap. § 1.4 interprets the production coefficient as a Kolmogorov-based thermostat. In Sec. VII we provide a physical interpretation of the geometry of uncertainty, before concluding in Sec. VIII.

Two-Field Framework

We consider an incompressible fluid described by a velocity field v(x,t)v(x,t) and a second scalar field s(x,t)s(x,t), which we interpret as the local intermittency or stochastic diffusivity associated with the microscopic path-uncertainty process introduced in our previous work [1].

Kinematics and invariants. The velocity gradient is decomposed into symmetric and antisymmetric parts,

∇v=S+Ω,(5)\nabla v=S+\Omega, \tag*{(5)}

with strain-rate tensor SS and rotation-rate tensor Ω\Omega given by

Sij=12(∂vi∂xj+∂vj∂xi),(6)S_{ij}=\frac{1}{2}\left(\frac{\partial v_i}{\partial x_j}+\frac{\partial v_j}{\partial x_i}\right), \tag*{(6)}
Ωij=12(∂vi∂xj−∂vj∂xi).(7)\Omega_{ij}=\frac{1}{2}\left(\frac{\partial v_i}{\partial x_j}-\frac{\partial v_j}{\partial x_i}\right). \tag*{(7)}

The vorticity is ω=∇×v\omega=\nabla\times v, and the scalar shear magnitude is

K:=2S:S.(8)K:=\sqrt{2S:S}. \tag*{(8)}

Momentum equation with emergent anisotropy. To close the momentum equation, we posit that the effective stress tensor is determined by the scalar magnitude ss but possesses a structure dictated by the local flow geometry. We retain the incompressible Navier–Stokes form for the left-hand side and interpret the new stress as the macroscopic footprint of the microscopic path uncertainty:

∂tv+(v⋅∇)v=−∇p+∇⋅Σ,∇⋅v=0,(9)\partial_t v+(v\cdot\nabla)v=-\nabla p+\nabla\cdot\Sigma,\qquad\nabla\cdot v=0, \tag*{(9)}

where the total viscous stress tensor Σ\Sigma is defined as:

Σ=2(ν0+s)S+τaniso.(10)\Sigma=2(\nu_0+s)S+\tau_{\mathrm{aniso}}. \tag*{(10)}

The term 2sS2sS represents an eddy-viscosity (isotropic) contribution to the deviatoric stress, consistent with identifying νeff=ν0+s\nu_{\mathrm{eff}}=\nu_0+s as the sum of molecular and turbulent (stochastic) diffusivities. The term τaniso\tau_{\mathrm{aniso}} captures stress anisotropy that emerges from the interaction between strain and rotation. Motivated by explicit algebraic Reynolds-stress modeling ideas, we define:

τaniso=C2sK(SΩ−ΩS).(11)\tau_{\mathrm{aniso}}=C_2\frac{s}{K}\left(S\Omega-\Omega S\right). \tag*{(11)}

Here, the commutator [S,Ω]=SΩ−ΩS[S,\Omega]=S\Omega-\Omega S quantifies the misalignment of principal stress and strain axes due to mean rotation. The coefficient s/Ks/K ensures dimensional consistency (L2/T2L^2/T^2) while linking the intensity of the anisotropy to the uncertainty field ss. In numerical implementation, the factor 1/K1/K requires regularization (e.g., 1/max⁡(K,ϵ)1/\max(K,\epsilon)) to avoid singularities in weak-strain regions.

We stress that Eq. (11) is a modeling choice (not derived from the Schrödinger-Bridge formalism) guided by objectivity and by the requirement that the anisotropic contribution vanish when s=0s=0.

The central question is how to define the evolution of ss in a principled way.

Mixed-Metric Gradient Flow for s

We consider the evolution of ss as a superposition of two gradient flows of an energy functional E[s]E[s], taken in different metrics.

Definition 1 (Mixed-metric gradient flow). Let

E[s]=Emorph[s]+Ereact[s],(12)E[s]=E_{\mathrm{morph}}[s]+E_{\mathrm{react}}[s], \tag*{(12)}
Emorph=∫12s2 dx,(13)E_{\mathrm{morph}}=\int\frac{1}{2}s^2\,\mathrm{d}x, \tag*{(13)}
Ereact=∫V(s;S,ω) dx.(14)E_{\mathrm{react}}=\int V(s;S,\omega)\,\mathrm{d}x. \tag*{(14)}

The evolution of ss is defined as the sum of:

  • the Wasserstein-2 gradient flow of EmorphE_{\mathrm{morph}} (morphology),

  • the L2L^2 gradient flow of EreactE_{\mathrm{react}} (reaction/coupling),

advected by the resolved velocity vv.

A clean intermediate step. Define

u:=12s2(u≥0).(15)u:=\frac{1}{2}s^2 \qquad(u\ge0). \tag*{(15)}

Note that

Δu=∇⋅(s∇s).(16)\Delta u=\nabla\cdot(s\nabla s). \tag*{(16)}

In particular, the Wasserstein-2 gradient flow of Emorph[s]=∫12s2 dxE_{\mathrm{morph}}[s]=\int\frac{1}{2}s^2\,\mathrm{d}x yields

(∂ts)morph=∇⋅(s∇δEmorphδs)=∇⋅(s∇s)=Δu,(17)(\partial_t s)_{\mathrm{morph}}=\nabla\cdot\left(s\nabla\frac{\delta E_{\mathrm{morph}}}{\delta s}\right)=\nabla\cdot(s\nabla s)=\Delta u, \tag*{(17)}

which makes explicit the porous-medium-type (degenerate) structure in the ss-variable.

Applying the mixed-metric construction leads to

∂ts+v⋅∇s=∇⋅(s∇s)⏟morphology+q(v,s)⏟coupling,(18)\partial_t s+v\cdot\nabla s=\underbrace{\nabla\cdot(s\nabla s)}_{\text{morphology}}+\underbrace{q(v,s)}_{\text{coupling}}, \tag*{(18)}

where the coupling term qq encodes exchange with the resolved flow through VV.

Steady balance and Poisson resolvent. In steady pockets where both the time derivative and advection are negligible relative to morphology and coupling, the equation reduces to

Δu=−q,(19)\Delta u=-q, \tag*{(19)}

with formal solution

u=(−Δ)−1q=∫0∞etΔq dt,(20)u=(-\Delta)^{-1}q=\int_0^\infty e^{t\Delta}q\,\mathrm{d}t, \tag*{(20)}

i.e. uu is obtained by convolving qq with the heat kernel. This resolvent structure mirrors the way the Dirichlet kernel organized the wall layer in our previous work [1].

Coupling term and dimensional consistency

We decompose

q=qprod−qrelax,(21)q=q_{\mathrm{prod}}-q_{\mathrm{relax},} \tag*{(21)}

where qprodq_{\mathrm{prod}} is a production term driven by three-dimensional vortex stretching, and qrelaxq_{\mathrm{relax}} represents relaxation driven by resolved shear.

We require qq to have dimension

[q]=L2T2,(22)[q]=\frac{L^2}{T^2}, \tag*{(22)}

so that all terms in Eq. (18) have consistent units. The basic quantities are summarized in Table I.

QuantitySymbolDimension
Intermittency (stochastic diffusivity)ssL2/TL^2/T
Shear magnitudeKKT−1T^{-1}
Strain-rate tensorSST−1T^{-1}
Vorticityω\omegaT−1T^{-1}
Production invariantI=∥Sω∥2\mathcal{I}=\lVert S\omega\rVert^2T−4T^{-4}
Linear relaxationsKsKL2/T2L^2/T^2
Quadratic relaxation (dimensionless C4C_4)(C4/ν0)s2K(C_4/\nu_0)s^2KL2/T2L^2/T^2
Production sourceCI(ν0+s)IC_I(\nu_0+s)\mathcal{I}L2/T2L^2/T^2

Table I. Dimensional consistency of quantities entering the ss-equation.

For clarity (and to keep free parameters dimensionless), we write the saturation coefficient as (C4/ν0)(C_4/\nu_0) with C4C_4 dimensionless, and we parametrize the production coefficient as CI=Ckτη3C_I=C_k\tau_\eta^3 with CkC_k dimensionless (see § 1.4).

Production invariant: vortex stretching and causal closure

A central modeling choice is the invariant driving production. We impose four requirements:

  1. Objectivity: it must be frame-indifferent, i.e. unaffected by superposed rigid-body motions [12, 13].

  1. Dimensionality: it must distinguish strictly two-dimensional limits (no vortex stretching) from genuinely three-dimensional turbulence [14].

  1. Consistency: it must have dimension T−4T^{-4} (Table I).

  1. Causality: it must be tied to vortex stretching/tilting, the kinematic mechanism behind the forward cascade [6, 15].

Kinematic requirement. Introduce the vorticity ω=∇×v\boldsymbol{\omega} = \nabla\times\boldsymbol{v} and the vortex-stretching vector w:=Sω\boldsymbol{w} := S\boldsymbol{\omega}. Guided by the vorticity-transport equation, we adopt:

Axiom 1 (Causal link). The local production rate qprodq_{\mathrm{prod}} must be proportional to the square of the vortex-stretching rate,

qprod∝(flow intensity)×∥Sω∥2,(23)q_{\mathrm{prod}} \propto\text{(flow intensity)} \times\lVert S\boldsymbol{\omega}\rVert^{2}, \tag*{(23)}

so that qprod≥0q_{\mathrm{prod}} \ge0 and production vanishes whenever there is no stretching or tilting of vorticity.

We choose

I(S,ω)=∥Sω∥2=(Sω)⋅(Sω),ω=∇×v.(24)\mathcal{I}(S,\boldsymbol{\omega}) = \lVert S\boldsymbol{\omega}\rVert^{2} = (S\boldsymbol{\omega}) \cdot(S\boldsymbol{\omega}), \qquad\boldsymbol{\omega} = \nabla\times\boldsymbol{v}. \tag*{(24)}

In any strictly two-dimensional flow v(x,y,t)=(u(x,y,t),v(x,y,t),0)\boldsymbol{v}(x,y,t) = (u(x,y,t),v(x,y,t),0) with no dependence on the third coordinate, the vorticity is purely out-of-plane and the symmetric strain cannot stretch it; one finds Sω=0S\boldsymbol{\omega} = 0 and therefore

I(S,ω)=0for all strictly two-dimensional flows.(25)\mathcal{I}(S,\boldsymbol{\omega}) = 0 \quad\text{for all strictly two-dimensional flows.} \tag*{(25)}

Dimensional closure. We model the source as

qprod=CI(ν0,s,ϵ)(ν0+s)I(S,ω),(26)q_{\mathrm{prod}} = C_{\mathcal{I}}(\nu_{0},s,\epsilon)(\nu_{0}+s)\mathcal{I}(S,\boldsymbol{\omega}), \tag*{(26)}

so (ν0+s)I(\nu_{0}+s)\mathcal{I} has dimension L2/T5L^{2}/T^{5} and therefore [CI]=T3[C_{\mathcal{I}}] = T^{3}.

Our previous work [1] motivates tying CIC_{\mathcal{I}} to the Kolmogorov time scale

τη=(ν0ϵ)1/2.(27)\tau_{\eta} = \left(\frac{\nu_{0}}{\epsilon}\right)^{1/2}. \tag*{(27)}

Axiom 2 (Dimensional closure). The coefficient CIC_{\mathcal{I}} is set by the local dissipation time scale, i.e. we take

CI(x,t)=Ckτη(x,t)3,(28)C_{\mathcal{I}}(x,t) = C_{k}\tau_{\eta}(x,t)^{3}, \tag*{(28)}

with CkC_{k} dimensionless.

Relaxation and saturation

Relaxation is modeled as

qrelax=sK+C4ν0s2K,(29)q_{\mathrm{relax}} = sK + \frac{C_{4}}{\nu_{0}}s^{2}K, \tag*{(29)}

where C4C_{4} is dimensionless. The linear term sKsK is dominant in the scale-invariant overlap layer, while the quadratic term provides saturation in regions where ss becomes large.

Collecting all contributions,

q(v,s)=CI(ν0,s,ϵ)(ν0+s)I(S,ω)−sK−C4ν0s2K.(30)q(\boldsymbol{v},s) = C_{\mathcal{I}}(\nu_{0},s,\epsilon)(\nu_{0}+s)\mathcal{I}(S,\boldsymbol{\omega}) - sK - \frac{C_{4}}{\nu_{0}}s^{2}K. \tag*{(30)}

Wall Asymptotics

We examine the behavior of Eq. (18) near a planar wall, focusing on two asymptotic regimes relevant in LES practice:

  • WRLES (wall-resolved LES): the near-wall region is sufficiently resolved that the modeled contribution must vanish smoothly at the wall;

  • WMLES (wall-modeled LES): the overlap is treated via an asymptotic model consistent with constant-stress arguments.

We consider a canonical plane channel with wall-normal coordinate yy, wall at y=0y=0, and mean velocity U(y)U(y) in the streamwise direction. We assume statistical stationarity and homogeneity in the streamwise and spanwise directions.

WRLES limit: near-wall behavior

In the viscous sublayer (y+≲10y^{+} \lesssim10), the mean flow is close to a simple shear. While instantaneous near-wall turbulence is three-dimensional, vortex stretching is strongly constrained; in the strict two-dimensional shear limit the invariant I\mathcal{I} vanishes identically. Thus, as a leading-order asymptotic model for the inner-most region, we take

qprod≈0,(31)q_{\mathrm{prod}} \approx0, \tag*{(31)}

and the steady ss-equation reduces to a balance between morphology and relaxation,

Δu≈qrelax,u=12s2.(32)\Delta u \approx q_{\mathrm{relax}}, \qquad u = \frac{1}{2}s^{2}. \tag*{(32)}

In a one-dimensional approximation (s=s(y)s=s(y), K≈K0K \approx K_{0} close to the wall), this reads

d2dy2(12s2)≈sK0+C4ν0s2K0.(33)\frac{\mathrm{d}^{2}}{\mathrm{d}y^{2}}\left(\frac{1}{2}s^{2}\right) \approx sK_{0} + \frac{C_{4}}{\nu_{0}}s^{2}K_{0}. \tag*{(33)}

Assuming no-slip and finite variance rate at the wall, we impose

s(0)=0,s(y)≥0,(s ∂ys)∣y=0=0,(34)s(0)=0,\qquad s(y)\ge0,\qquad\left.(s\,\partial_y s)\right|_{y=0}=0, \tag*{(34)}

where the last condition is the no-flux morphology boundary condition.

A slightly more robust leading-order argument. Assume a smooth expansion s(y)=a1y+a2y2+O(y3)s(y)=a_1y+a_2y^2+O(y^3). The no-flux condition ss′∣y=0=0ss'|_{y=0}=0 allows a1a_1 but does not enforce it. Substituting into (33) shows that if a1≠0a_1\ne0 then the left-hand side behaves like a constant while the right-hand side behaves like O(y)O(y), which is inconsistent. Hence a1=0a_1=0 and the first nonzero term is quadratic.

Equivalently, seeking a power law s(y)∼cyns(y)\sim cy^n as y→0y\to0 yields the same conclusion. With u=12s2∼12c2y2nu=\frac{1}{2}s^2\sim\frac{1}{2}c^2y^{2n} one has u′′∼12c2(2n)(2n−1)y2n−2u''\sim\frac{1}{2}c^2(2n)(2n-1)y^{2n-2}, whereas the leading relaxation term is sK0∼cK0ynsK_0\sim cK_0y^n. Balancing exponents gives 2n−2=n2n-2=n, i.e. n=2n=2:

s(y)∼cy2,u(y)∼12c2y4as y→0.(35)s(y)\sim cy^2,\qquad u(y)\sim\frac{1}{2}c^2y^4\qquad\text{as }y\to0. \tag*{(35)}

Connection to practical WRLES damping and the y3y^3 condition. In many wall-resolved LES implementations based on Smagorinsky-type or dynamic SGS models, the eddy viscosity is multiplied by a van-Driest damping function so that the modeled turbulent viscosity behaves as

νtSGS(y+)∝(y+)3as y+→0,(36)\nu_t^{\mathrm{SGS}}(y^+)\propto(y^+)^3\qquad\text{as }y^+\to0, \tag*{(36)}

while recovering νt∝y\nu_t\propto y in the logarithmic region; see, for instance, Härtel & Kleiser [16] and van Driest [17]. In our framework, the intrinsic small-yy scaling is s∼y2s\sim y^2, i.e. νtcore(y)=s(y)∼y2\nu_t^{\mathrm{core}}(y)=s(y)\sim y^2 in the resolved limit.

If one wishes to enforce νt∼y3\nu_t\sim y^3 at very small y+y^+ for coarse grids, this can be achieved by composing ss with a damping function D(y+)D(y^+), for example

νtWRLES(y)=D(y+)s(y),D(y+)∼y+(y+→0),(37)\nu_t^{\mathrm{WRLES}}(y)=D(y^+)s(y),\qquad D(y^+)\sim y^+\quad(y^+\to0), \tag*{(37)}

so s∼y2s\sim y^2 together with D∼y+D\sim y^+ yields νtWRLES∼y3\nu_t^{\mathrm{WRLES}}\sim y^3 without modifying the core two-field PDE.

WMLES limit: overlap layer and logarithmic law

In the overlap layer, we assume:

  1. A constant-stress region: τxy≈ρuτ2\tau_{xy}\approx\rho u_\tau^2.

  2. High Reynolds number: s≫ν0s\gg\nu_0, so νeff≈s\nu_{\mathrm{eff}}\approx s.

  3. Overlap-layer scale invariance: no length scale other than yy enters the leading-order balance [18, 19, 20].

Mean-momentum balance. In a steady, fully developed channel,

(ν0+s)U′≈uτ2,(38)(\nu_0+s)U'\approx u_\tau^2, \tag*{(38)}

where U′=dU/dyU'=dU/dy. The anisotropic stress term (11) does not modify the mean shear-stress balance for parallel shear flow.

In the overlap limit ν0≪s\nu_0\ll s,

s(y)U′(y)≈uτ2.(39)s(y)U'(y)\approx u_\tau^2. \tag*{(39)}

Scale invariance selects s(y)∝ys(y)\propto y. To determine the yy-dependence of ss, assume that the leading-order overlap dynamics introduce no length scale other than yy. Consider the one-dimensional morphology operator,

L[s]:=ddy(s(y)ddy).(40)\mathcal{L}[s]:=\frac{d}{dy}\left(s(y)\frac{d}{dy}\right). \tag*{(40)}

If s(y)∼yαs(y)\sim y^\alpha, then under dilation y↦λyy\mapsto\lambda y,

L[s](λy)∼λα−2L[s](y).(41)\mathcal{L}[s](\lambda y)\sim\lambda^{\alpha-2}\mathcal{L}[s](y). \tag*{(41)}

Overlap-layer scale invariance (no intrinsic length beyond yy) selects α=1\alpha=1, hence

s(y)=cy,(42)s(y)=cy, \tag*{(42)}

with cc a constant with units of velocity.

Substituting (42) into (39) yields

cyU′(y)≈uτ2⇒U′(y)≈uτ2c1y,(43)cyU'(y)\approx u_\tau^2\quad\Rightarrow\quad U'(y)\approx\frac{u_\tau^2}{c}\frac{1}{y}, \tag*{(43)}

so

U(y)≈uτ2cln⁡y+B.(44)U(y)\approx\frac{u_\tau^2}{c}\ln y+B. \tag*{(44)}

Identifying c=κuτc=\kappa u_\tau, we recover

U+(y+)=1κln⁡y++B+.(45)U^+(y^+)=\frac{1}{\kappa}\ln y^++B^+. \tag*{(45)}

Theorem 1 (Logarithmic law from scale-invariant ss). Under the assumptions of constant stress, high Reynolds number (ν0≪s\nu_0\ll s), and overlap-layer scale invariance, the only nontrivial overlap-layer profile satisfies s(y)∝ys(y)\propto y, yielding a logarithmic mean velocity profile.

Energetic Interpretation and the CIC_I Thermostat

The production coefficient CIC_I in Eq. (26) is interpreted energetically. Define the local dissipation rate per unit mass by

ϵ(x,t)=νeffK2.(46)\epsilon(x,t)=\nu_{\mathrm{eff}}K^2. \tag*{(46)}

In our previous work [1] we motivated the Kolmogorov time scale

τη=(ν0ϵ)1/2(47)\tau_{\eta} = \left(\frac{\nu_0}{\epsilon}\right)^{1/2} \tag*{(47)}

as a dissipation-controlled microscopic adjustment time. A minimal choice consistent with [CI]=T3[C_I] = T^3 is

CI(x,t)=Ckτη3=Ck(ν0ϵ(x,t))3/2,(48)C_I(x,t) = C_k\tau_{\eta}^3 = C_k\left(\frac{\nu_0}{\epsilon(x,t)}\right)^{3/2}, \tag*{(48)}

with CkC_k dimensionless. This makes production strongest where stretching is intense and dissipation time scales are long, and weaker where dissipation is already strong.

Relation to Classical Turbulence Models

Eddy viscosity and mixing length. In Smagorinsky-type LES [3, 4], one writes

νt=(CSΔ)2K.(49)\nu_t = (C_S\Delta)^2 K. \tag*{(49)}

In mixing-length theories one recovers νt∝y\nu_t \propto y in the overlap to obtain the log law [6]. Here νeff=ν0+s\nu_{\mathrm{eff}} = \nu_0 + s is not imposed algebraically but emerges from (18); in the overlap we obtain s∝ys \propto y from scale invariance, recovering the mixing-length behavior without postulating it.

Two-equation models and anisotropy. In kk–ϵ\epsilon or kk–ω\omega models [7, 8], νt\nu_t is related to independently evolved turbulence quantities. The present framework evolves only one scalar field ss, but includes a commutator-driven anisotropic correction (11) to capture a minimal rotation-induced misalignment of stress and strain.

WRLES vs WMLES. The WRLES and WMLES limits appear naturally as asymptotic regimes of the same PDE. In the inner region, production is strongly constrained (and vanishes in the strict 2D limit), while in the overlap region scale invariance selects s∝ys \propto y. Reviews and modern perspectives on wall modeling and WMLES can be found in [21, 22], and unified wall-resolved/wall-modeled strategies include, e.g., [23].

RANS one-equation baselines. For baseline context, one-equation RANS closures such as Spalart–Allmaras [24] and hybrid RANS/LES strategies such as DES [25] provide useful reference points. The present two-field setting is not intended as a drop-in replacement, but as an alternative geometric organizing principle.

Complex geometries and immersed boundary methods. In complex geometries, wall modeling is often combined with immersed boundary methods and related sharp-interface techniques. While discretization is outside our scope, standard references include Peskin [26] and Mittal & Iaccarino [27]. For pressure-gradient boundary-layer context relevant to wall-turbulence modeling, see also [28].

The Geometry of Uncertainty: A Physical Interpretation

The ontology of ss (magnitude). The porous-medium morphology term, ∇⋅(s∇s)\nabla\cdot(s\nabla s), supports finite-speed spreading of compactly supported disturbances. In this sense the uncertainty field organizes as a propagating front rather than diffusing instantaneously as in a linear heat equation.

The emergence of anisotropy (structure). The nonlinear stress closure τaniso∝sκ[S,Ω]\tau_{\mathrm{aniso}} \propto\frac{s}{\kappa}[S,\Omega] represents a minimal geometric frustration: strain tends to align stress, while rotation induces reorientation. The commutator measures the failure of simultaneous diagonalization of SS and Ω\Omega, and thus provides an objective signature of rotation-induced misalignment.

Causality and the forward cascade. Driving production via I=∥Sω∥2\mathcal{I} = \lVert S\omega\rVert^2 enforces a kinematic condition: sustained production occurs only where vortex stretching is active, and ceases in strictly two-dimensional limits.

Conclusion

Building on our previous work [1], we developed a two-field framework in which the velocity vv and an intermittency field ss evolve in a coupled manner. The effective viscosity is νeff=ν0+s\nu_{\mathrm{eff}} = \nu_0 + s, and the total stress includes an emergent commutator term sκ[S,Ω]\frac{s}{\kappa}[S,\Omega] that breaks pure Boussinesq isotropy.

In a wall-resolved asymptotic limit, constraining production through the vortex-stretching invariant and balancing morphology with relaxation yields s∼y2s \sim y^2 as y→0y \to0. In the overlap layer, high Reynolds number, constant stress and scale invariance yield s(y)∝ys(y) \propto y as the unique nontrivial profile, recovering the logarithmic law of the wall.

Acknowledgments

M.S.-A. acknowledges financial support from the EU Doctoral Network MODELAIR. M.S.-A. thanks Emelie Saga Stark for numerous discussions and insightful suggestions. Special thanks to Francesco Mario D’Aferio and Eduardo Terres-Caballero for discussions that sharpened the conceptual foundations.

Derivation of anomalous scaling and K62 intermittency corrections

This appendix shows how the multiplicative structure of the ss-dynamics yields log-normal statistics for coarse-grained dissipation and reproduces the Kolmogorov–Obukhov refined similarity hypothesis (K62) [29, 30, 31].

Multiplicative dynamics along Lagrangian trajectories

Define the material derivative Dt:=∂t+v⋅∇D_t := \partial_t + \boldsymbol{v} \cdot\nabla. In the high-Reynolds-number limit (ν0≪s\nu_0 \ll s), and on local time scales for which the morphology term is subdominant relative to source/sink terms, Eq. (18) reduces to

Dts≈qprod−qrelax.(50)D_t s \approx q_{\mathrm{prod}} - q_{\mathrm{relax}}. \tag*{(50)}

Using Eqs. (26)–(29), dropping ν0\nu_0 against ss, and retaining the linear relaxation (dominant in overlap/inertial-range balances), we obtain

Dts≈s(CII(S,ω)−K).(51)D_t s \approx s\left(C_I\mathcal{I}(S,\omega)-K\right). \tag*{(51)}

Defining the net amplification rate

ξ(t):=CI(t)I(t)−K(t),(52)\xi(t) := C_I(t)\mathcal{I}(t)-K(t), \tag*{(52)}

we get the multiplicative (logarithmic) law

Dtln⁡s=ξ(t).(53)D_t\ln s=\xi(t). \tag*{(53)}

The quadratic saturation term provides nonlinear damping at very large ss; Eq. (53) remains the leading-order driver of multiplicative fluctuations when linear relaxation dominates.

Log-normality of ss

Integrating Eq. (53) along a trajectory from t=0t=0 to t=Tt=T,

ln⁡s(T)s(0)=∫0Tξ(t′) dt′.(54)\ln\frac{s(T)}{s(0)}=\int_0^T \xi(t')\,\mathrm{d}t'. \tag*{(54)}

If ξ(t)\xi(t) is stationary with finite correlation time and satisfies a central-limit-type condition for TT sufficiently larger than that correlation time, the integral approaches a Gaussian random variable. Hence ln⁡s\ln s is approximately normal and ss is approximately log-normally distributed.

Coarse-grained dissipation and K62 variance

For ν0≪s\nu_0\ll s,

ϵ(x,t)=νeffK2≈sK2.(55)\epsilon(x,t)=\nu_{\mathrm{eff}}K^2\approx sK^2. \tag*{(55)}

Define coarse-grained dissipation over a ball of size rr,

ϵr(x,t):=1∣Br∣∫Br(x)ϵ(y,t) dy.(56)\epsilon_r(x,t):=\frac{1}{\lvert B_r\rvert}\int_{B_r(x)}\epsilon(y,t)\,\mathrm{d}y. \tag*{(56)}

K62 assumes that the cascade depth scales like ln⁡(L/r)\ln(L/r), giving

Var⁡(ln⁡ϵr)=μln⁡(Lr),(57)\operatorname{Var}(\ln\epsilon_r)=\mu\ln\left(\frac{L}{r}\right), \tag*{(57)}

with intermittency coefficient μ\mu.

Choosing the mean of ln⁡ϵr\ln\epsilon_r to enforce ⟨ϵr⟩=ϵ0\langle\epsilon_r\rangle=\epsilon_0 yields the log-normal moment formula

⟨ϵrq⟩=ϵ0q(Lr)μ2q(q−1).(58)\left\langle\epsilon_r^q\right\rangle=\epsilon_0^q\left(\frac{L}{r}\right)^{\frac{\mu}{2}q(q-1)}. \tag*{(58)}

Anomalous scaling of structure functions

Refined similarity gives (in the inertial range)

⟨∣δur∣p⟩∼⟨ϵrp/3⟩rp/3.(59)\left\langle\lvert\delta u_r\rvert^p\right\rangle\sim\left\langle\epsilon_r^{p/3}\right\rangle r^{p/3}. \tag*{(59)}

Taking q=p/3q=p/3 in Eq. (58) gives

⟨∣δur∣p⟩∼rζp,ζp=p3−μ18p(p−3),(60)\left\langle\lvert\delta u_r\rvert^p\right\rangle\sim r^{\zeta_p},\qquad\zeta_p=\frac{p}{3}-\frac{\mu}{18}p(p-3), \tag*{(60)}

which is the standard K62 log-normal intermittency correction.

References

References

  1. [1]M. Sanchi-Agudo and R. Vinuesa, A geometric foundation for the universal laws of turbulence, arXiv:2512.00068 [physics.flu-dyn].
  2. [2]V. I. Arnold, Sur la géométrie différentielle des groupes de Lie de dimension infinie et ses applications à l’hydrodynamique des fluides parfaits, Ann. Inst. Fourier 16, 319 (1966).
  3. [3]J. Smagorinsky, General circulation experiments with the primitive equations: I. The basic experiment, Mon. Weather Rev. 91, 99 (1963).
  4. [4]D. K. Lilly, The representation of small-scale turbulence in numerical simulation experiments, Proc. IBM Sci. Comput. Symp. Environ. Sci. , 195 (1967).DOI
  5. [5]U. Schumann, Subgrid scale model for finite difference simulations of turbulent flows in plane channels and annuli, J. Comput. Phys. 18, 376 (1975).
  6. [6]S. B. Pope, Turbulent Flows (Cambridge University Press, 2000).
  7. [7]B. E. Launder and D. B. Spalding, The numerical computation of turbulent flows, Comput. Methods Appl. Mech. Eng. 3, 269 (1974).
  8. [8]F. R. Menter, Two-equation eddy-viscosity turbulence models for engineering applications, AIAA J. 32, 1598 (1994).DOI
  9. [9]R. Jordan, D. Kinderlehrer, and F. Otto, The variational formulation of the Fokker–Planck equation, SIAM J. Math. Anal. 29, 1 (1998).
  10. [10]F. Otto, The geometry of dissipative evolution equations: The porous medium equation, Commun. Partial Differ. Equ. 26, 101 (2001).DOI
  11. [11]J. L. Vázquez, The Porous Medium Equation: Mathematical Theory (Oxford University Press, 2007).
  12. [12]C. Truesdell and W. Noll, The Non-Linear Field Theories of Mechanics, Handbuch der Physik, Vol. III/3 (Springer-Verlag, Berlin, 1965).
  13. [13]C. G. Speziale, On the advantages of the vorticity transport equation in turbulence modeling, Journal of Computational Physics 73, 476 (1987).
  14. [14]R. H. Kraichnan, Inertial ranges in two-dimensional turbulence, The Physics of Fluids 10, 1417 (1967).DOI
  15. [15]H. Tennekes and J. L. Lumley, A First Course in Turbulence (MIT Press, Cambridge, MA, 1972).
  16. [16]C. Härtel and L. Kleiser, Analysis and modelling of subgrid-scale motions in near-wall turbulence, Journal of Fluid Mechanics 356, 327 (1998).DOI
  17. [17]E. R. van Driest, On turbulent flow near a wall, Journal of the Aeronautical Sciences 23, 1007 (1956).
  18. [18]A. A. Townsend, The Structure of Turbulent Shear Flow, 2nd ed. (Cambridge University Press, 1976).
  19. [19]I. Marusic, J. P. Monty, M. Hultmark, and A. J. Smits, On the logarithmic region in wall turbulence, J. Fluid Mech. 716, R3 (2013).DOI
  20. [20]S. Hoyas, R. Vinuesa, P. J. Schmid, and H. M. Nagib, Sensitivity study of the logarithmic velocity profile in wall-bounded turbulence, Phys. Rev. Fluids 9, 054601 (2024).
  21. [21]U. Piomelli and E. Balaras, Wall-layer models for large-eddy simulations, Annual Review of Fluid Mechanics 34, 349 (2002).
  22. [22]S. T. Bose and G. I. Park, Wall-modeled large-eddy simulation for complex turbulent flows, Annual Review of Fluid Mechanics 50, 535 (2018).DOI
  23. [23]F. De Vanna, M. Cogo, M. Bernardini, F. Picano, and E. Benini, A unified wall-resolved and wall-modeled method for large-eddy simulation of compressible wall-bounded flows, Phys. Rev. Fluids 6, 034614 (2021).DOI
  24. [24]P. R. Spalart and S. R. Allmaras, A one-equation turbulence model for aerodynamic flows, AIAA Paper 10.2514/6.1992-439 (1992).DOI
  25. [25]P. R. Spalart, Detached-eddy simulation, Annual Review of Fluid Mechanics 41, 181 (2009).DOI
  26. [26]C. S. Peskin, The immersed boundary method, Acta Numerica 11, 479 (2002).
  27. [27]R. Mittal and G. Iaccarino, Immersed boundary methods, Annual Review of Fluid Mechanics 37, 239 (2005).
  28. [28]R. Vinuesa, P. H. Rozier, P. Schlatter, and H. M. Nagib, Experiments and computations of localized pressure gradients with different history effects, AIAA Journal 52, 368 (2014).
  29. [29]A. N. Kolmogorov, A refinement of previous hypotheses concerning the local structure of turbulence in a viscous incompressible fluid at high reynolds number, Journal of Fluid Mechanics 13, 82 (1962).DOI
  30. [30]A. M. Obukhov, Some specific features of atmospheric turbulence, Journal of Fluid Mechanics 13, 77 (1962).DOI
  31. [31]U. Frisch, Turbulence: The Legacy of A. N. Kolmogorov (Cambridge University Press, 1995).

Paper details

vlnc:W_x5jqhxaq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents